wrap-up
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill makes extensive use of shell commands to automate git operations, including
git status,git commit,git push, andgit pull. It also usestaskkillon Windows to forcibly terminategit.exeprocesses if a push operation hangs due to credential prompts. While these are necessary for the skill's function, they represent a broad command execution surface.- [INDIRECT_PROMPT_INJECTION]: In Phase 2, the skill is instructed to scan the conversation history for errors and pending actions. This session-derived data is then interpolated into reports, commit messages, and 'Continuation Handoff' files (Phase 6). This creates an attack surface where malicious content previously introduced into the session could influence the agent's automated outputs and next-session instructions. - Ingestion point: Conversation history scan in Phase 2 (Checks 3 and 4).
- Boundary markers: Not used for interpolated session data.
- Capability inventory: File writes (handoffs, logs), git commits, and git pushes.
- Sanitization: Absent; session summaries are directly used in prompts and messages.- [DATA_EXFILTRATION]: The skill automatically performs
git pushto remote repositories. While this is its primary purpose, it constitutes an automated data export capability. The skill attempts to mitigate risk via 'Gate 3', which runs a filename scan and a content-aware privacy scanner (scripts/scan_staged_privacy.py) to detect PII like IBANs and national ID numbers before pushing to public forges.- [DYNAMIC_EXECUTION]: The skill executes multiple local Python scripts at runtime, includingtests/run_tests.py,scripts/check_desktop_sync.py,scripts/scan_staged_privacy.py, andscripts/handoff_status.py. These scripts are executed via subprocess calls with parameters derived from the local environment.- [EXTERNAL_DOWNLOADS]: Phase 5 includes a 'Skills-lock' loop that executesnpx skills addto synchronize or update core agent skills from a remote source (gasserane/personal-skills), representing the installation of external code at the end of a session.
Audit Metadata